Pavel Dubovik comments

Results 18 comments of


                                            Pavel Dubovik

Bug with tweedie loss training on GPU

And is it solution? Need to set up leaf_estimation_iterations=20?

Bug with tweedie loss training on GPU

> Thank you. I confirm that setting `leaf_estimation_iterations=20` produces approximately same results for GPU and CPU on `CatBoost 1.2.7`. In the same time, the parameter `leaf_estimation_iterations=20` slows down the estimation...

Bug with tweedie loss training on GPU

@Evgueni-Petrov-aka-espetrov , @andrey-khropov Any update please?

Incorrect display of metrics for model.eval_metrics or when training with custom_metric with tweedie or poisson loss function

Hi guys! Any updates? Это очень простой баг который скорее всего легко исправить) Но он безумно глупый и часто вводит в заблуждение

Could we redirect pr_bar to a new logger like from tqdm.contrib.logging import logging_redirect_tqdm?

could you elaborate on the idea what functionality exactly would you like to see?

Can't specify params when running inference on MLServer

Hello @svpino! Any updates here?

Can't specify params when running inference on MLServer

Is there any hope that you will get around to this in the near future?

Can't specify params when running inference on MLServer

@svpino But this seems like a major oversight. The mlflow documentation recommends using MLServer for industrial deployment of ML models. And here it turns out that you can't use parameters...